Journal of Hydrology 261 (2002) 86±101

www.elsevier.com/locate/jhydrol

Intraseasonal dynamics of soil moisture variability within a small

agricultural maize cropped ®eld
F. Hupet*, M. Vanclooster
Department of Environmental Sciences and Land Use Planning, Universite Catholique de Louvain, Unite de Genie Rural,
Place Croix du Sud 2, BP 2, B-1348 Louvain-la-Neuve, Belgium
Received 22 May 2001; revised 13 November 2001; accepted 19 December 2001

Abstract
The spatio-temporal dynamics of soil water content was investigated within a small agricultural maize cropped ®eld located
in Belgium. Soil moisture measurements were intensively made between May 30 and September 13, 1999 on 28 sampling
locations at different depths (from 0 to 125 cm) with both TDR and neutron probe. The adopted sampling scheme resulted in a
comprehensive data set of nearly 8000 soil moisture measurements. Using this data set, we ®rst probe the role of factors
controlling the spatio-temporal soil moisture dynamics for the different considered soil depths. Special emphasis is thereby
given to the role of the vegetation in the space±time relationships of soil moisture. Secondly, we identify the temporal dynamics
of the spatial structure of soil moisture patterns at different soil depths. Thirdly, we investigate the relationship between the
mean soil moisture and the spatial variability across time, analysing through the season the optimal sampling strategies to adopt
for providing the ®eld areal soil moisture within a given prede®ned error limit. The results showed that the vegetation, spatially
variable within the ®eld, and subsequently through the process of evapotranspiration and the root water uptake, plays a non-
negligible role in the temporal dynamics of the observed soil moisture patterns for the super®cial layers. The spatial structure of
these soil moisture patterns was non-existent or only weakly marked. The study ®nally indicated that a negative correlation
exists between the spatial variability and the mean soil moisture, implicitly suggesting that the sampling has to be more
intensive for the drier conditions. Besides these results, this study reemphasises the importance of conducting soil moisture
spatial variability studies with measurements performed on the entire hydrological active zone and to adopt temporally
unchanged sampling locations in order to progress in the thorough understanding of the physical processes generating the
soil moisture spatial variability. q 2002 Elsevier Science B.V. All rights reserved.
Keywords: Soil moisture; Spatial variability; Field-scale; Vegetation growth; Sampling

1. Introduction fall into in®ltration and runoff and the partitioning of

Soil moisture is a key state variable for a wide vari-
ety of hydrological processes acting over a range of
spatial and temporal scales. For example, soil moisture
is one of the factors in¯uencing the partitioning of rain-
* Corresponding author.
E-mail addresses: hupet@geru.ucl.ac.be (F. Hupet),
vanclooster@geru.ucl.ac.be (M. Vanclooster).
0022-1694/02/$ - see front matter q 2002 Elsevier Science B.V. All rights reserved.
PII: S 0022-169 4(02)00016-1
net radiation into sensible and latent heat. In addition, it
controls the subsurface drainage of water and thereby
the leaching of chemicals to the groundwater. It
is therefore evident that a thorough understanding of
the soil moisture behaviour is of primary importance
for soil hydrological research and engineering applica-
tions in areas such as irrigation scheduling, soil
based precision farming, hydrological modelling,
¯ood forecasting, and estimating the groundwater
 F. Hupet, M. Vanclooster / Journal of Hydrology 261 (2002) 86±101
recharge. During the last decade an increasing number
of studies have been published focusing on the areal
estimation of soil moisture, considering explicitly the
spatio-temporal variations of this property (e.g. Crave
and Gascuel-Odoux, 1997; Grayson and Western,
1998; Famiglietti et al., 1999). These studies, often
based on the near-surface soil moisture (0±5 cm),
were conducted at different spatial scales (1 m 2 to a
few km 2), at different temporal scales (few days to
few years), with different measurement techniques
(e.g. gravimetric sampling, TDR, remote sensing), in
a variety of hydrologic and climatic conditions. Within
all these studies, the important spatio-temporal
dynamics of soil moisture have been put forward.
This spatio-temporal variability is in¯uenced by topo-
graphic features such as the soil surface slope angle
(Hills and Reynolds, 1969; Moore et al., 1988; Nyberg,
1996) and slope orientation (Reid, 1973; Western et al.,
1999a), by the soil (hydrodynamics) properties
(Henninger et al., 1976; Crave and Gascuel-Odoux,
1997), by the vegetation distribution (Bouten et al.,
1992; Mohanty et al., 2000a), by land use and in parti-
cular the agricultural practices (Famiglietti et al.,
1999), by the climatic variability and by the combined
effects of some of those factors (e.g. Hawley et al.,
1983). However, until now the environmental control-
ling factors have been investigated rather unequally.
For example, topographical control of soil moisture
has been studied far more extensively than that of
vegetation. This research bias can mainly be explained
by the current availability of detailed digital elevation
models, which can be used to derive statistical distri-
butions of terrain indices related to the spatial patterns
of soil moisture. The vegetation factor on the other
hand is, in numerous studies conducted at the ®eld-
scale, explicitly considered as uniform across the
studied area (e.g. Famiglietti et al., 1998; Mohanty et
al., 2000b). However, everyone carefully examining
an agricultural ®eld or forest stand can easily discern
a variable vegetation growth. This variable growth
integrates the effect of numerous interacting factors
such as the climatic environment (e.g. radiation in¯ux),
the stress (disease attack, soil moisture and nutrient
content), the soil properties and some possible human
disturbances. Furthermore, it is generally well
accepted that the vegetation growth is in one way or
another related to the transpired ¯uxes (e.g. Doorenbos
and Kassam, 1987). In this context, we can suspect
87
that during the vegetative period, the vegetation
through the process of evapotranspiration plays a
non-negligible role in the development of spatial soil
moisture patterns within ®elds. This role of the evapo-
transpiration process was already implicitly suggested
by Grayson et al. (1997) within its concepts of
preferred states and local control in the development
of spatial soil moisture patterns and by Western et al.
(1999a) with the very small-scale variations contribut-
ing signi®cantly to the poor performance of the terrain
indices within the Tarrawarra catchment.
The observed soil moisture variability within a ®eld
is likely to re¯ect the combined in¯uences of the
above-mentioned factors acting at different spatial
scales. In order to quantify and describe spatial varia-
bility of soil moisture, the geostatistical correlation
structure needs to be identi®ed. Indeed, quantitative
estimations of this structure are required for a number
of purposes such as, for example at the ®eld-scale, the
interpolation of spatial patterns from point measure-
ments and the subsequent estimation of the average
®eld soil moisture, and for example the determination
of the optimal spacing between sampling points.
Nevertheless, rigorous quantitative estimations of
the geostatistical structure of soil moisture are prone
to various dif®culties as described by Western et al.
(1998) and Western and BloÈschl (1999) as the
observed geostatistical structure is directly a function
of the soil moisture measurement scale de®ned by the
spacing, extent and support.
Besides the challenge of determining the relative
roles of the different factors governing the spatio-
temporal dynamics of soil moisture and of deriving
the geostatistical structure of the soil moisture
patterns, more in-depth knowledge on the relationship
between the ®eld-scale means and standard deviations
of soil moisture across time is needed. Whereas some
authors have found a positive correlation between the
standard deviation and the mean soil moisture (Hills
and Reynolds, 1969; Reynolds, 1970; Bell et al.,
1980; Robinson and Dean, 1993; Famiglietti et al.,
1998) other investigations found no good predictive
relationship (Hawley et al., 1983; Charpentier and
Groffman, 1992). Some others studies (e.g. Fami-
glietti et al., 1999) showed relationships where the
variance is negatively correlated to the mean soil
moisture. Until now, and in the light of the diversity
of previous results, there is no well-established
88 F. Hupet, M. Vanclooster / Journal of Hydrology 261 (2002) 86±101

Fig. 1. The experimental ®eld, Louvain-la-Neuve, Belgium. Topographic features of the ®eld (1 ha) are shown with the sampling locations
indicated with circle (studied area ˆ 0.63 ha).

consensus about this relationship whereas a sound
knowledge of this is of a primary importance for a
number of purposes. For example, such relationship
can be interesting to pronounce about the underlying
soil moisture spatial variability within remote
sensing pixels or to determine strategic sampling
periods which minimise the errors for a given
sample size.
Although there have been a series of studies
which have improved our understanding of the
dynamics of the soil moisture variability, additional
studies for different spatial and temporal scales and
for various hydroclimatic and geographic settings
are needed (Georgakakos and Baumer, 1996).
Keeping in mind that for many hydrologic applica-
tions the whole soil moisture pro®le is of important
signi®cance (Western et al., 1998), soil moisture
variability should be studied for the whole soil
pro®le (or at least over a depth of several deci-
meters) rather than just over a thin layer at the
soil surface (BloÈschl and Sivapalan, 1995). In this
framework, we conducted an experiment at the
®eld-scale, initially designed to investigate the in
situ spatial variability of evapotranspiration, provid-
ing a comprehensive set of soil moisture data,
allowing us to tackle some of the above-discussed
issues. The main objectives of this study are:
1. to analyse the temporal dynamics of the soil
moisture spatial variability and to probe the
 F. Hupet, M. Vanclooster / Journal of Hydrology 261 (2002) 86±101
role of governing factors for the different consid-
ered soil depths. Special emphasis will be given
to the role of the vegetation in the space±time
relationships of soil moisture;
2. to identify the temporal dynamics of the spatial
structure of soil moisture patterns at different
soil depths; and
3. to investigate the relationship between the mean
soil moisture and the spatial variability across
time, analysing through the season the optimal
sampling strategies (in terms of number of
samples) to adopt for providing the ®eld areal
soil moisture within a given prede®ned error
limit.
2. Material and methods
2.1. Experimental site and measurements
The experiment was conducted in a 10,000 m 2 agri-
cultural ®eld located in Louvain-La-Neuve (Belgium)
on a plateau. The study was focused within an area of
6300 m 2 centred on the ®eld. A contour map of the
®eld (Fig. 1) elucidates that the surface relief is rather
uniform with a maximum relief of 3 m for the whole
®eld and only 1.8 m for the area of interest. A weak
topographic convergence is present between sampling
points 22 and 23 (Fig. 1). Because of its speci®c topo-
graphic position, the experimental ®eld is assumed not
to be in¯uenced by the adjacent areas through the
process of lateral surface water ¯ow. The soil is clas-
si®ed as a well-drained silty-loam (Aba, according to
the Belgian soil classi®cation). Textural analysis for
the super®cial layer (0±40 cm) yield a loam content of
82%, a clay content of 12% and a sand content of 6%.
Fig. 1 also shows the adopted regular sampling grid
of 28 point measurement, each separated by 15 m
sampling intervals. The sampling points were located
by means of a Differential Global Positioning System
(DGPS). On each sampling point, we installed verti-
cally a triple wire TDR probe to monitor the soil
moisture in the top 20 cm of the soil pro®le, an
150 cm access tube to measure with a neutron probe
soil moisture at 25, 50, 75, 100 and 125 cm depths and
two tensiometers at 115 and 135 cm (data not used in
this study). Between the 30th of May and the 13th of
89
September 1999, soil moisture measurements were
made on 60 dates by means of TDR (cable tester
Tektronix 1502 B) and on 45 dates with a neutron
probe device (Institute of Hydrology Neutron Probe
System II). This resulted in a comprehensive data set
of nearly 8000 soil moisture measurements.
Climate at the site is moderate humid with an aver-
age annual rainfall of 780 mm (Dupriez and Sneyers,
1982), which is relatively well distributed over the
year and with an average annual ETo of 527.6 mm
(Gellens-Meulenberghs and Gellens, 1992). A
weather station located near the ®eld provided all
climatic variables, i.e. dry and wet temperatures,
shortwave radiation, wind speed and rainfall. The
course of the experiment was marked by three long
low rainfall periods. The ®rst occurred between the
7th and the 20th of June (DOY 158±171) with a rain-
fall amount of only 1.9 mm, the second between the
6th of July and the 3th of August (DOY 187±215)
with a rainfall amount of 8 mm, and the third between
the 20th of August and the 13th of September (DOY
232±256) with a rainfall amount of 1.6 mm. These
periods were interrupted by rainy events, some of
them with very high intensities (i.e. DOY 215).
The ®eld was cropped with maize (Zea mays L.)
which was sown on the 25th of April 1999 with a
density of 110,000 plants per hectare and harvested
on the 10th of October 1999 (DOY 283). The rows
were spaced 75 cm apart and all the measurement
instruments were installed in the middle of the rows.
On each of the 28 sampling locations, we selected 10
maize plants (5 within each row located on each side
of the sampling location) that were considered as
those in¯uencing, through the process of root water
uptake, the measured soil moisture pro®le. Note that
these 10 maize plants are therefore situated on the
border of a 75 £ 75 cm 2 in the centre of which is
measured the soil moisture pro®le. On each of the
280 selected maize plants, we measured the number
of growing leaves as well as the number and size of
full grown leaves, thereby deriving the mean leaf area
index (LAI) on the 12th of July and the 5th of Septem-
ber for each 28 sites according to the methodology
proposed by Ruget et al. (1996). The ®nal yield was
also measured on each 28 sampling points on the 17th
September 1999.
During the experimental campaign and the follow-
ing winter, two in situ calibration curves were
90 F. Hupet, M. Vanclooster / Journal of Hydrology 261 (2002) 86±101

Fig. 2. Daily precipitation and space±time ®elds of volumetric soil moisture content for the different depths: (a) 0±20 cm, (b) 25 cm, (c) 50 cm,
(d) 75 cm, (e) 100 cm, (f) 125 cm.

established, respectively, for the TDR and neutron
probe. For each calibration curve, 25 pairs of observed
vs reference soil moisture were measured within the
range 20±45%. The gravimetric method was consid-
ered as the reference method. To derive the total
variance associated with an individual neutron probe
measurement, we used the methodology proposed by
Haverkamp et al. (1984) and Vauclin et al. (1984).
This methodology was extended (not presented
here) for the TDR measurements. The resulting total
 F. Hupet, M. Vanclooster / Journal of Hydrology 261 (2002) 86±101
variance (i.e. the instrument and calibration compo-
nents) associated with an individual soil moisture
estimation is 1.2 and 1.8 (%v/v) 2, respectively, for
the neutron probe and for TDR. Consequently, the
respective con®dence intervals (probability level of
95%) for an estimated soil moisture content are
^2.24 and ^2.75%.
2.2. Data analysis
First of all, we investigated systematically for each
depth and each sampling day if the soil moisture data
were normally distributed both with the visual inspec-
tion of normal probability plots and with the Shapiro±
Wilk statistic (Shapiro and Wilk, 1965). The degree of
spatial dependence was then investigated by using
standard geostatistical techniques (Isaaks and Srivas-
tava, 1989) such as the visual examination of sample
semivariograms. The sample semivariogram, gs …h† at
a given lag, h, is given by
1 X
gs …h† ˆ … u 2 uj † 2 …1†
2N…h† i;j i
with N the number of pairs of observations for a speci-
®ed lag, ui and uj the soil moisture at sampling loca-
tions, respectively, i and j and the summation
performed for all the i, j pairs of observations of the
selected lag. In order to ensure a minimum number of
pairs of observations within each lag, only omnisdir-
ectional semivariograms were used with a minimum
and maximum lag of, respectively, 15 and 58 m. This
resulted in a minimum of 33 soil moisture data pairs
for each selected lag interval, except 20 and 23 pairs
for the 42.5 and 45 m lags. The semivariograms were
only plotted for the normally or close to normally
distributed soil moisture values. Geostatistical analy-
sis was done by means of a geostatistical toolbox
programmed in a MATLABe environment.
For quantifying the number of samples necessary to
estimate the mean with a given accuracy for a given
probability level, we applied the central limit theorem
ta2 s2
nˆ …
d within the region but also within ®elds (Dudal, 1953).
where n is the number of samples, s 2 is the estimated
sample variance, ta2 the Student's parameter corre-
sponding to the probability level and d is the relative
accuracy. For the soil moisture values which were not
91
normally distributed, the results obtained by the
central limit theorem were checked by applying the
bootstrapping technique as used by Dane et al. (1986)
with 1000 bootstrap replicates.
3. Results and discussion
3.1. Spatio-temporal dynamics of the soil moisture
Space±time ®elds of volumetric soil moisture
content for the different depths are given in Fig. 2
on which the Y-axis numbering is referring to the
sampling locations indicated in Fig. 1. Also shown
are the daily precipitation depths. A thorough inspec-
tion of Fig. 2 allows details of the spatio-temporal
behaviour of soil moisture within the ®eld to be
addressed. First of all, the soil moisture temporal
dynamics is obviously depth dependent. The temporal
dynamics is more pronounced closer to the soil
surface, where soil is subjected to the root water
uptake and rainfall events. The three low rainfall peri-
ods are relatively well identi®ed for the super®cial soil
layers, especially for the second drying period (DOY
187±215). Indeed, as a result of the important tran-
spiration from the well developed maize crop
combined with a high climatic demand, the water
depletion of the super®cial layers (0±20, 25 cm) is
abundant during the second drying period. For the
third period, the climatic demand is quite similar,
yet the maize crop much more senescent, resulting
in a less depletion. The temporal dynamics in the
intermediate layers (50, 75 and 100 cm) is clearly
more attenuated. Nevertheless, these layers become
progressively drier in response to the root water
uptake combined with the upward ¯ux. Fig. 2 also
shows that, within the deeper layers (75, 100 and
125 cm), the temporal persistence of some dry
(sampling locations 4, 5 and 16) and wet zones (loca-
tion 23) within the ®eld remains high. The persistence
of drier zones can be explained by the sandy material
underlying the loamy soil. Indeed, it is well known
2† that the thickness of the loamy layer is quite variable
It is therefore assumed that the `dry' sampling loca-
tions (4, 5 and 16) are situated where the loamy soil is
shallow, leading to measuring the soil moisture
within the transition zone between the loamy soil
92 F. Hupet, M. Vanclooster / Journal of Hydrology 261 (2002) 86±101

Fig. 3. Daily precipitation depths and intraseasonal evolution of the spatial variability and the mean soil moisture content for the differents soil
depths.

and the sandy material. Visual observations of the soil
pro®le on the sampling locations 4 and 5 and some
u 2 h ®eld values for these locations con®rm this
hypothesis. The sampling location 23 on the other
hand is very wet at 75, 100 and 125 cm depth. It is
hypothesised that a ®ne textured layer at this location
leads locally to a perched water table.
The qualititative information on the spatial
variability of soil moisture within the ®eld from Fig.
2 is quantitatively summarised in Fig. 3. This ®gure
shows the temporal evolution of the observed spatial
variances expressed in terms of standard deviation
and the mean soil moisture for the different depths.
Also shown are the daily precipitation depths for a
better insight of the temporal dynamics of the spatial
variability.
 F. Hupet, M. Vanclooster / Journal of Hydrology 261 (2002) 86±101 93

Table 1 20 cm is due to the smaller sampling volume of

Mean LAI of 10 plants measured on each sampling location (Sl) on TDR. For the deeper layers (100 and 125 cm depth),
DOY 193. Also shown are the yields (dry matter production)
measured on DOY 260
the observed spatial variability increases again with a
value for the maximum standard deviation of 0.04 and
Sl LAI Yield (g) Sl LAI Yield (g) 0.042, respectively. This high variability observed at
these depths is almost exclusively explained by some
1 2.81 1690 15 3.97 1923
2 3.34 2028 16 3.82 2086 singular wet and dry spots deeper in the pro®le. As
3 3.44 1767 17 3.47 1644 explained earlier, those spots can probably be related
4 3.26 1944 18 3.35 1887 to areas of different texture, i.e. different hydraulic
5 4.26 2755 19 3.99 2267 properties, leading to a variable observed soil moist-
6 2.62 1537 20 4.09 2262
ure. Note that those observed values of soil moisture
7 4.21 2038 21 3.71 2039
8 3.41 1680 22 3.58 1744 spatial variability are in the range of those found in
9 4.32 2425 23 3.94 2087 other ®elds and small catchment-scale experiments
10 3.85 2199 24 3.25 1751 (e.g. Charpentier and Groffman, 1992; Western et
11 3.65 2136 25 3.67 2228 al., 1999b). Note also that it was considered in our
12 4.05 2446 26 3.74 2208
study that all the observed soil moisture variance is
13 4.12 2397 27 3.71 2264
14 3.77 2163 28 2.22 1552 explained by the spatial variability within the ®eld.
Indeed, whereas the measurement variances (i.e. the
calibration and instrument components) on an indivi-
First of all, Fig. 3 reveals that the temporal varia- dual soil moisture estimation for neutron probe and
tions of the mean daily soil moisture is depth depen- TDR, respectively, are 1.2 and 1.8 (%v/v) 2, these
dent, and less pronounced deeper in the pro®le. variances are drastically reduced for an areal soil
Indeed, the observed range for the 0±20 cm layer is moisture estimation. This was effectively con®rmed
between 0.4 and 0.21, for the 25 cm 0.38±0.22, for (results not shown in this paper) with the methodology
the 50 cm 0.36±0.24, for the 75 cm 0.37±0.28, for the developed by Vauclin et al. (1984) for the neutron
100 cm 0.36±0.31 and for the 125 cm 0.35± probe and with a similar methodology developed for
0.31 cm 3 cm 23. Similar results, but for longer time the TDR.
periods, were already observed by Loague (1992). In the following paragraphs we explain the
The observed spatial variability is also depth depen- observed temporal dynamics of the spatial soil moist-
dent. For the 0±75 cm layer which can be considered ure patterns. Firstly, it can be easily observed in Fig. 3
as the active root-zone, the spatial variability tends to that the temporal stability of the spatial variability is
decrease with depth with a maximum standard devia- more pronounced for the deeper soil layers, whereas
tion of 0.038 for the 0±20 cm, 0.035 for the 25 cm for the active root-zone (0±75 cm depth) the spatial
depth, 0.031 for the 50 cm depth and 0.03 for the variability is much more time variant. For example, a
75 cm depth. In this context, one should remember sharp increase of the soil moisture spatial variability is
that two different techniques, TDR and neutron observed during the second dry period (DOY 187±
probe, were used to measure soil water content, 215). This increase is well marked for the 0±20 and
each with speci®c sampling volumes estimated, 25 cm depths and also with a time lag for the 50 and
respectively, at approximately 1000 and 15,000 cm 3. 75 cm depths. As during such a dry period the
Although it is virtually impossible to quanti®y the temporal dynamics of the soil moisture is mainly
direct impact of the two different sampling volumes governed by the root water uptake and as the LAI
on the observed variability, this might explain in some measured on DOY 193 on each sampling location is
extent the larger variability of the ®rst 20 cm (with indeed very variable (see Table 1) we can suspect that
TDR) compared to that measured at other depths (with this increase in the spatial variability is mainly driven
neutron probe). Nevertheless, as observed variability by a variable root water uptake within the ®eld. To test
increases gradually from 75 to 25 cm for neutron this hypothesis, the root water uptake, here assimi-
probe measurements, there is no reasons to believe lated to the soil water depletion, between DOY 192
that the larger variability observed for the ®rst and 215 was determined on each sampling location by
94 F. Hupet, M. Vanclooster / Journal of Hydrology 261 (2002) 86±101

Fig. 4. Relation between the LAI and the soil water depletion (mm) for the different soil depths for the period between 09/07 and 02/08 (DOY
192±215).

means of a simpli®ed water balance …uDOY192 2
uDOY215 † for the different depths. As shown in Fig.
4, the calculated root water uptake was then correlated
with the measured LAI. Although these correlations
are not well de®ned (linear regression with maximum
r 2 of 0.31 for 25 and 50 cm depths), there is a trend
between the root water uptake and the LAI especially
for the 25, 50 and 75 cm depths. That tends to con®rm
that the root water uptake is spatially variable within
the ®eld at a given depth, partially due to the variable
areal crop growth, and that it partially controls the
temporal dynamics of the spatial soil moisture varia-
bility. Furthermore, as shown in Fig. 3, the increasing
spatial variability is delayed for the deeper layer as the
root water uptake ®rstly affects the surface soil layers.
Indeed, the spatial variability increases substantially
from DOY 190 onwards for the 25 cm depth, DOY
205 for the 50 cm depth and only from DOY 210 for
the 75 cm depth. All these observations con®rm the
important role of the vegetation in explaining soil
moisture variability, as was already suggested by
Grayson et al. (1997) and Western et al. (1999a).
We showed that vegetation is spatially variable within
the ®eld, inducing variable evapotranspiration rates
(not yet quanti®ed) and consequently variable root
water uptake rates. It is often put forward that the
variable radiation in¯ux is the main factor controlling
the spatial variability of evapotranspiration. Yet, it is
 F. Hupet, M. Vanclooster / Journal of Hydrology 261 (2002) 86±101 95

Fig. 5. Sample omnisdirectional semivariograms calculated for DOY 214 for the six different depths: (a) 0±20 cm, (b) 25 cm, (c) 50 cm, (d)
75 cm, (e) 100 cm, (f) 125 cm. Also shown are the observed total variances (dotted line).

not relevant to consider this source of variation as
explanatory factor, given the scale of our ®eld study
and the nearly ¯at topography.
After the long low rainfall period (DOY 187±215),
the rainfall event of DOY 215 tends to slightly decrease
the spatial variability of the 0±20 cm layer. This
decrease can probably be explained by the high intensity
of this rainfall event (20 mm in less than 20 min.)
combined with some variable hydraulic soil properties
leading to some differential in®ltration in the topsoil.
After this rainfall event, the spatial variability increases
again for the 0±20 and 25 cm layers until DOY 232 from
which it slightly decreases. If we consider that the
increasing soil moisture spatial variability during the
period DOY 160±230 is mainly explained by the vari-
able root water uptake, vegetation can explain about 6.7
(%v/v) 2 of the variance for the 0±20 cm depth and 6.2
(%v/v) 2 of the variance for the 25 cm depth. In light of
such results, it is not surprising that terrain indices will in
some cases perform poorly in explaining soil moisture
variability as was already suggested by Western et al.
(1999a). It is important to note that, despite the lack of
quantitative information about the spatial variability of
soil hydraulic properties of the surface layers, these
properties must probably also contribute partially to
the soil moisture variability. For the two deepest depths
(100 and 125 cm), we observe that the high spatial varia-
bility is very temporally stable and slightly increasing
96 F. Hupet, M. Vanclooster / Journal of Hydrology 261 (2002) 86±101
Fig. 6. Standard deviation of soil moisture content (cm 3 cm 23) versus mean soil moisture content (cm 3 cm 23) for the different soil depths.
during the experiment. As a matter of fact, it is not so
surprising because the main factors controlling the
temporal dynamics of the soil moisture variability at
this scale are acting mainly on the super®cial layers.
3.2. Spatial structure
In a ®rst step, we investigated the normality of the
soil moisture data for each day and depth. The
Shapiro±Wilk statistic and the visual inspection of
the probability plots showed that the data are
normally or approximately normally distributed for
the surface soil layers (0±20, 25, 50 and 75 cm). For
the two deepest soil depths (100 and 125 cm), the
data are non-normal but tending towards normality.
Omnisdirectional sample semivariograms were then
calculated systematically for each depth but only for
the sampling days where the data were normal or
close to normality. Fig. 5 shows the sample semivar-
iograms calculated for each depth for one represen-
tative run (DOY 215). Also shown in Fig. 5(f) is the
number of pairs of observations for the different lags.
No apparent spatial structure was found for that day
(DOY 215) and semivariograms calculated for other
days showed similar behaviour. Nevertheless, for the
two deeper depths, there is a general increasing trend
in semi-variance with increasing lag, suggesting that
the sill is not reached for the maximum lag. In fact,
results (not shown here) of a more recent experiment
within the same ®eld with 50 sampling locations at
0.9 m intervals suggest that there is a well-de®ned
anisotropic spatial correlation of 40 m for the 100
and 125 cm depths. Unfortunately, the experimental
set-up of 1999 did not allow us to investigate this
anisotropy. For the surface layers (0±75 cm) no
spatial structure can be discerned considering the
adopted sampling scheme. These results are in agree-
ment with the ®ndings of Charpentier and Groffman
(1992) who carried out a similar experiment in terms
of measurement extent and spacing. The observed
lack of spatial structure for the super®cial soil layers
is not so surprising if we consider that the vegetation
is effectively the main factor controlling the devel-
opment of the soil moisture patterns. Indeed as the
vegetation growth is variable at short distance (few
meters), we can suspect that spatial correlations are
existent but for smaller lags than those investigated.
Furthermore, our results are consistent with the ®nd-
ings of Musters and Bouten (1999) showing a range
of 14 m in a forest stand.
 F. Hupet, M. Vanclooster / Journal of Hydrology 261 (2002) 86±101 97

Fig. 7. Intraseasonal evolution of the number of samples for three selected soil depths (0±20 cm and b, 50 cm c and d and 125 cm depth e and f)
for 95 and 99% probability level for different levels of absolute error (^1.5± ^ 2.5± ^ 3.5% volumetric soil moisture content).

3.3. Sampling analysis
The soil moisture data set collected in this study
allows investigation of the relationship between the
®eld-scale means and variances of soil moisture
content across time. If such relationships exist, they
can be of great signi®cance, for example, to optimise
the number of samples required to estimate the mean
value within a speci®ed limit of error. Fig. 6 shows the
standard deviation of soil moisture within the ®eld
versus its mean value for the different depths. We
can observe that the absolute variability, i.e. the stan-
dard deviation, increase quite linearly with decreasing
mean soil moisture. Furthermore, Fig. 6 shows that
this relationship is depth dependent, with important
differences between the super®cial depths (0±75 cm)
and more profound depths (100 and 125 cm). This
negative correlation is consistent with the previous
98 F. Hupet, M. Vanclooster / Journal of Hydrology 261 (2002) 86±101

Table 2
Summary of the number of samples (minimum and maximum) needed to reach a given absolute precision for the different depths for two given
probability levels. Results ( p ) are also presented for 100 cm depth without `outliers' (sampling points 4, 5 and 23) and 125 cm depth (sampling
points 4, 5, 16 and 23)

Depth (cm) 95% Probability level 99% Probability level

^ 1.5% ^ 2.5% ^ 3.5% ^ 1.5% ^ 2.5% ^ 3.5%

0±20 2±33 1±12 1±6 5±60 1±22 1±11

25 5±23 1±8 1±4 6±43 3±15 1±7
50 2±17 1±6 1±3 5±31 1±11 1±5
75 4±17 1±6 1±3 7±30 2±11 1±5
100 13±32 4±11 2±5 23±59 8±21 5±10
100 p 1±7 1±2 1±2 1±14 1±5 1±2
125 20±33 7±11 3±6 34±60 12±21 6±11
125 p 1±8 1±2 1±2 3±13 1±4 1±2

®ndings of Famiglietti et al. (1999), who analysed the
surface soil moisture within six different cultivated
relatively ¯at remote sensing pixels (0.64 km 2).
Nevertheless, they are in contrast with other previous
investigations (e.g. Hills and Reynolds, 1969; Hennin-
ger et al., 1976; Famiglietti et al., 1998). Note most of
these studies, in contrast with our results, were
conducted on experimental sites with quite well
pronounced topographic features.
The observed relationship between the ®eld-scale
means and standard deviations of soil moisture
directly affects the number of samples required to
reach a same level of accuracy on the ®eld-scale
mean soil moisture. The time series of the number
of required samples was calculated for each depth
with the central limit theorem. As the W statistic
(Shapiro and Wilk, 1965) showed that the soil moist-
ure content was not systematically normally distribu-
ted over the ®eld, especially for the deeper layers (100
and 125 cm), we con®rmed the results obtained by
means of the central limit theorem with those of the
bootstrapping method. Comparisons showed very
small differences between the two techniques. Fig. 7
shows the intraseasonal evolution of the number of
samples for three selected depths (0±20 cm a and b,
50 cm c and d and 125 cm depth e and f) for two
probability levels (95 and 99%) and for three absolute
errors (1.5±2.5±3.5% volumetric soil moisture).
Firstly, we can observe in Fig. 7 that the number of
samples used in our experiment was suf®cient to esti-
mate the mean within an absolute error of 1.5% for a
probability level of 95%. A striking feature is the
temporal variations of the number of samples, which
is depth dependent. This is a corollary of the temporal
behaviour of the soil moisture spatial variability
presented in Fig. 3. For the 99% probability level,
we show that the number of samples drastically
increases. This ®gure also illustrates quite clearly
that optimal sampling strategies have to be adjusted
across the season. Optimising the sampling design
could also be investigated by the use of some techni-
ques such as the one based on the time stability
concept (e.g. Vachaud et al., 1985; Grayson and
Western, 1998). Although this technique has not yet
been investigated in our study, Fig. 2 suggests that
this kind of technique could be appealing mainly
for the deeper layers for which a strong time stability
is observed.
Table 2 summarises the minimum and maximum
number of samples for the course of our experiment.
These are relatively low for the estimations of the
mean soil moisture within an absolute error of 3.5%,
slightly increasing for an absolute error of 2.5% and
becoming large, considering the small sampled area,
for the 1.5% error. It is also clear in Table 2 that the
number of required samples decreases from the soil
surface to the 75 cm depth to increase again for the
deeper layer. Nevertheless, if the same exercise is
done for 100 and 125 cm depths without `outliers'
(sampling locations 4, 5, 16 and 23), the number of
samples is strongly reduced (cf. Table 2). The
increasing soil moisture spatial variability between
75 cm depth and the soil surface also suggests that
the number of samples required for estimating
 F. Hupet, M. Vanclooster / Journal of Hydrology 261 (2002) 86±101
Fig. 8. Relation between the mean soil moisture content and the number of samples required to reach the mean ^1.5 at the 95% probability
level.
appropriately the mean of the 0±5 cm soil moisture
content, which is of interest for teledetection
purposes, could increase even further. Finally, Fig. 8
shows the relationship between the mean ®eld soil
moisture and the number of samples needed to reach
the mean ^1.5 at the 95% probability level. Such
relationships, which have to be con®rmed in the
next few years with other investigations, could be of
the greatest interest for the hydrological community
dealing with ground truthing of remote sensing data,
estimation of mean soil moisture in hydrological
modelling, irrigation scheduling and other soil
hydrological applications.
4. Conclusions
In this paper, we addressed three basic questions
related to the spatial (®eld-scale) and temporal (intra-
seasonal) dynamics of the soil moisture content.
These questions were addressed by means of a
comprehensive data set of nearly 8000 soil moisture
measurements, collected on 28 locations at different
depths (from 0 to 125 cm) within a small cropped
®eld. The ®rst question deals with the factors control-
ling the spatial soil moisture patterns. We observed
99
during the vegetative growth period that the soil
moisture spatial variability tends to strongly increase
in the super®cial soil layers (0±50 cm). In order to
elucidate this behaviour, we quanti®ed the root
water uptake based on a simpli®ed water balance.
We showed that the calculated root water uptake
variability, like the measured vegetation growth,
was important. Furthermore, signi®cant correlations
between the vegetation growth and root water uptake
was observed. In view of these observations, we
conclude that the vegetation growth, and subsequently
the evapotranspiration and the root water uptake play
a non-negligible role in the temporal dynamics of the
observed soil moisture patterns. Considering that the
spatial variability is due to the root water uptake, vari-
able vegetation growth could explain spatial variance
of 6.2±6.7 (%v/v) 2 for the super®cial layers. Such
high values can also explain the poor performance
of the terrain indices as shown, among others, by
Western et al. (1999a). To the authors' knowledge,
there have been no previous detailed studies
conducted at the ®eld-scale investigating the role of
the vegetation in the spatio-temporal dynamics of the
soil moisture. In this context, this study aimed to bring
new insights into the role of the vegetation until now
often put forward but never quanti®ed.
100 F. Hupet, M. Vanclooster / Journal of Hydrology 261 (2002) 86±101
The second question was related to the soil moist-
ure spatial structure. We showed that for the adopted
sampling scheme, the observed spatial structure was
non-existent or only weakly marked. For this kind of
small ®elds, without signi®cant relief, and during the
vegetative period where vegetation would act as the
main controlling factor, it is quite logical to observe
absence of signi®cant spatial structure for the super-
®cial soil layers.
The third question addresses the sampling problem
for estimating mean soil moisture. The study indicated
that a negative correlation exists between the spatial
variability and mean soil moisture. This relationship
implicitly suggests that the sampling has to be more
intensive for the drier conditions. For an absolute
error of 2.5 (%v/v) and for a probability level of
95%, the number of samples required is quite low
varying between 1 and 12 according to the depth
and time. On the other hand, this number increases
to a maximum of 33 for an absolute error of 1.5 (%v/
v). One should realise that all these above-presented
conclusions are based on data from just one season
and one location, further from a small ®eld, so their
generality will need to be tested with results from
other locations.
Finally, the authors stress again the importance to
conduct soil moisture spatial variability studies with
measurements performed on the entire hydrologically
active zone and not only on the ®rst few centimeters
of the soil. As the small-scale variability is often
important, we also recommend adoption of tempo-
rally unchanged sampling locations. Following these
two basic recommendations, we believe progress in
the thorough understanding of the physical processes
generating the soil moisture spatial variability can be
made. Finally, future studies focusing on the soil
moisture controlling factors (e.g. topographic
features) must explicitly and rigorously take into
account the within ®eld spatial variability of the
vegetation.
Acknowledgements
The authors thank E. Auquiere and C. Lucau for
their assistance with the use of the DGPS and P.
Bogaert for the free use of the geostatistical toolboxes.
We wish to thank G. Rentmeesters for the design and
installation of the experimental set-up. Farmer N.
Braibant is also acknowledged for providing free
access to the ®eld and his willing cooperation during
all the course of the experiment.
References
Bell, K.R., Blanchard, B.J., Schmugge, T.J., Witczak, M.W., 1980.
Analysis of surface soil moisture variations within large-®eld
sites. Water Resources Research 16 (4), 796±810.
BloÈschl, G., Sivapalan, M., 1995. Scale issues in hydrological
modelling: a review. In: Kalma, J.D., Sivapalan, M. (Eds.).
Advances in Hydrological Processes, Scale Issues in Hydrolo-
gical Modelling. Wiley, Chicester, pp. 9±48.
Bouten, W., Heimovaara, T.J., Tiktak, A., 1992. Spatial patterns of
throughfall and soil water dynamics in a Douglas ®r stand.
Water Resources Research 28 (12), 3227±3233.
Charpentier, M.A., Groffman, P., 1992. Soil moisture variability
within remote sensing pixels. Journal of Geophysical Research
97, 18987±18995.
Crave, A., Gascuel-Odoux, C., 1997. The in¯uence of topography
on time and space distribution of soil surface water content.
Hydrological Processes 11, 203±210.
Dane, J.H., Reed, R.B., Hopmans, J.W., 1986. Estimating soil para-
meters and sample size by bootstrapping. Soil Science Society
of America Journal 50, 283±287.
Doorenbos, J., Kassam, A.H., 1987. ReÂponse des rendements aÁ
l'eau. Bulletin FAO d'irrigation et de drainage, 235 Papier,
33, Rome.
Dudal, R., 1953. Etude morphologique et geÂneÂtique d'une seÂquence
de sols sur limon loessique. Agricultura I (2), 119±163.
Dupriez, G.L., Sneyers, R., 1982. PreÂcipitations: les normales du
reÂseau pluviomeÂtrique belge. Institut Royal MeÂteÂorologique de
Belgique, Publication seÂrie A, no. 101.
Famiglietti, J.S., Rudnicki, J.W., Rodell, M., 1998. Variability in
surface moisture content along a hillslope transect: Rattlesnake
Hill, Texas. Journal of Hydrology 210, 259±281.
Famiglietti, J.S., Devereaux, J.A., Laymon, C.A., Tsegaye, T.,
Houser, P.R., Jackson, T.J., Graham, S.T., Rodell, M., van
Oevelen, P.J., 1999. Ground-based investigation of soil moist-
ure variability within remote sensing footprints during the
Southern Great Plains 1997 (SGP97) Hydrology Experiment.
Water Resources Research 35 (6), 1839±1851.
Gellens-Meulenberghs, F., Gellens, D., 1992. L'eÂvapotranspiration
potentielle en Belgique: variabilite spatiale et temporelle.
Institut Royal MeÂteÂorologique de Belgique, Publication seÂrie
A, no. 130.
Georgakakos, K.P., Baumer, O.W., 1996. Measurement and utiliza-
tion of on-site soil moisture data. Journal of Hydrology 184,
131±152.
Grayson, R.B., Western, A.W., 1998. Towards areal estimation of
soil water content from point measurements: time and space
stability of mean response. Journal of Hydrology 207, 68±82.
Grayson, R.B., Western, A.W., Chiew, H.S., BloÈschl, G., 1997.
Preferred states in spatial soil moisture patterns: local and
 F. Hupet, M. Vanclooster / Journal of Hydrology 261 (2002) 86±101
nonlocal controls. Water Resources Research 33 (12), 2897±
2908.
Haverkamp, R., Vauclin, M., Vachaud, G., 1984. Error analysis in
estimating soil water content from neutron probe measurements:
1. Local standpoint. Soil Science 137, 78±90.
Hawley, M., Jackson, T., McCuen, R., 1983. Surface soil moisture
variation on small agricultural watersheds. Journal of Hydrol-
ogy 62, 179±200.
Henninger, D.L., Petersen, G.W., Engman, E.T., 1976. Surface soil
moisture within a watershed-variations, factors in¯uencing, and
relationship to surface runoff. Soil Science Society of America
Journal 40, 773±776.
Hills, R.C., Reynolds, S.G., 1969. Illustrations of soil moisture
variability in selected areas and plots of different sizes. Journal
of Hydrology 8, 27±47.
Isaaks, E.H., Srivastava, R.M., 1989. An Introduction to Applied
Geostatistics. Oxford University Press, New York pp. 561.
Loague, K., 1992. Soil water content at R-5. Part 1. Spatial and
temporal variability. Journal of Hydrology 139, 233±251.
Mohanty, B.P., Famiglietti, J.S., Skaggs, T.H., 2000a. Evolution of
soil moisture spatial structure in a mixed vegetation pixel during
the Southern Great Plains 1997 (SGP97) Hydrology Experi-
ment. Water Resources Research 36 (12), 3675±3687.
Mohanty, B.P., Skaggs, T.H., Famiglietti, J.S., 2000b. Analysis and
mapping of ®eld-scale soil moisture variability using high-
resolution, ground-based data during the Southern Great Plains
1997 (SGP97) Hydrology Experiment. Water Resources
Research 36 (4), 1023±1031.
Moore, I.D., Burch, G.J., Mackenzie, D.H., 1988. Topographic
effects on the distribution of surface soil water and the location
of ephemeral gullies. Transactions of the ASAE 31 (4), 1098±
1107.
Musters, P.A.D., Bouten, W., 1999. Assessing rooting depths of an
Austrian pine stand by inverse modeling soil water content
maps. Water Resources Research 35 (10), 3041±3048.
Nyberg, L., 1996. Spatial variability of soil water content in the
101
covered catchment at GardsjoÈn, Sweden. Hydrological
Processes 10, 89±103.
Reid, I., 1973. The in¯uence of slope orientation upon the soil
moisture regime, and its hydrogeomorphological signi®cance.
Journal of Hydrology 19, 309±321.
Reynolds, S.G., 1970. The gravimetric method of soil moisture
determination. Part III: an examination of factors in¯uencing
soil moisture variability. Journal of Hydrology 11, 288±300.
Robinson, M., Dean, T.J., 1993. Measurement of near surface soil
water content using a capacitance probe. Hydrological
Processes 7, 77±86.
Ruget, F., Bonhomme, R., Chartier, M., 1996. Estimation simple de
la surface foliaire de plantes de maõÈs en croissance. Agronomie
16, 553±562.
Shapiro, S.S., Wilk, M.B., 1965. An analysis of variance test for
normality. Biometrika 52, 591±611.
Vachaud, G., Passerat de Silans, A., Balabanis, P., Vauclin, M.,
1985. Temporal stability of spatially measured soil water prob-
ability density function. Soil Science Society of America Jour-
nal 49, 822±828.
Vauclin, M., Haverkamp, R., Vachaud, G., 1984. Error analysis in
estimating soil water content from neutron probe measurements:
2. Spatial standpoint. Soil Science 137, 141±148.
Western, A.W., BloÈschl, G., 1999. On the spatial scaling of soil
moisture. Journal of Hydrology 205, 20±37.
Western, A.W., BloÈschl, G., Grayson, R.B., 1998. Geostatistical
characterisation of soil moisture patterns in the Tarrawarra
catchment. Journal of Hydrology 205, 20±37.
Western, A.W., Grayson, R.B., BloÈschl, G., Willgoose, G.R.,
McMahon, T.A., 1999a. Observed spatial organization of soil
moisture and its relation to terrain indices. Water Resources
Research 35 (3), 797±810.
Western, A.W., Grayson, R.B., Green, T.R., 1999b. The Tarrawarra
project: high resolution spatial measurement, modelling and
analysis of soil moisture and hydrological response. Hydro-
logical Processes 13, 633±652.